Development of multi-model ensemble climate estimates using open-source environment

Abstract

We demonstrate an application of an R open source environment to climatological problems, namely, to the construction of multi-model ensemble estimates. The R ecosystem is used to develop a computation code which automates an ensembling procedure. We apply the code to the CMIP5 runoff data in the Russian territory for a preliminary analysis of the long-term natural variability. It has been found that a multi-model projection demonstrates a non-monotonic behaviour which contradicts the classic "wet getting wetter" pattern, but perfectly corresponds to the long-term natural variability. A detailed analysis of reproducibility of the runoff temporary structure seems to be unavoidable for successful forecasting of the long-term runoff trends.